Tuesday 14th July
Wednesday 15th July
Thursday 16th July
Friday 17th July
Wednesday 15th July, 11:00 - 12:30 Room: O-201
Measurement errors in the wealth surveys 1
|| Mr Junyi
Zhu (Deutsche Bundesbank )
|Coordinator 1||John Sabelhaus (Federal Reserve Board)|
|Coordinator 2||Brian Bucks (Consumer Financial Protection Bureau)|
Obtaining a comprehensive picture of households’ balance sheets and understanding their wealth accumulation process is of increasing interest to a large audience ranging from poli-cymakers and researchers to the general public. Consequently, more and more wealth surveys have been established worldwide. However, wealth data are susceptible to measurement errors specific to the nature of various asset and liability items. For example, households may not assess the value and amount of their assets constantly. And the valuation of less traded or distinctive assets is not straightforward. The knowledge required to answer some question can be demanding. Financial topics are always sensitive. Typically, questions on ownership of assets or liabilities are answered more accurately than questions on their value and in most cases the reporting quality of the debts outperforms that of the assets. Households from both ends of the wealth distribution are hard to identify and reach. The longitudinal data adds another layer of difficulty in distinguishing true changes from measurement errors. On the other hand, reporting error, the main measurement error, does not have a homogeneous pattern but can be classified.
We would like to invite survey practitioners to discuss how to detect and tackle measurement errors in wealth surveys. Researchers can analyze the missing pattern within the survey as a signal of potential errors. Matching to external surveys or administrative data and utilizing the panel dimension are other options to gauge the plausibility of answers. But then, there have been many prevention and reconciling measures. They include careful design and sequencing of questions, specialized interviewer training, software real-time checks, editing by reviewing the comments, dependent interviewing, etc. Innovative approaches are especially welcome. For example, using tax records, property lien data, online finance websites or other sources can fill the gap in building comprehensive profile of wealth accumulation.
Paper Details1. Asking highly flexible forms of disaggregated income and single total question: a case study from PHF on measuring income in general purpose household survey
Mr Junyi Zhu
The PHF questionnaire allows the respondent to answer income question in flexible and disaggregated alternatives: components, individual and household levels, accounting period, gross or net, quantity in brackets and different currencies. We have developed a tax microsimulation model to facilitate the standardization and imputation. The survey also collects a total monthly household disposable income as many general purpose surveys do. This paper first documents our microsimulation model by emphasizing this innovation in design and implementation. We then evaluate the data quality and potential improvement from this innovation which serves a guideline for other general purpose surveys.
2. The Mechanisms of Item Nonresponse and Measurement Error in Income Questions
Ms Barbara Felderer
(University of Mannheim)
Income questions usually suffer from high rates of item nonresponse. While imputation strategies assume that the propensity for nonresponse does not depend on a person's income, it is not clear whether this assumption is realistic.
3. Comparing range questions vs. unfolding brackets as a means of reducing non-response on financial questions
Even for respondents who do report their income, not much is known of the quality of these reports. If income is reported with non-random error, analyses will be biased.
Linking survey income information to individual register data allows for investigation the relationship between a person's income and item nonresponse and measurement error. These relationships will be compared across web and telephone mode.
Dr Mary Beth Ofstedal
(Survey Research Center, University of Michigan)
Dr Mick Couper (Survey Research Center, University of Michigan)
This paper reports results from an experiment in the 2011 U.S. Health and Retirement Study Internet Survey (n=4,500 persons age 51+) that was designed to evaluate data quality for two non-response followup approaches for financial questions. Respondents who did not provide an exact amount on any of three income sources were randomized to receive either a range question or a series of unfolding brackets as a followup. Findings suggest that unfolding brackets result in slightly higher item-missing rates than the range question for all three income sources and that the income distributions based on the